Ding, Yi2023-01-042023-01-042022-10https://hdl.handle.net/11299/250419University of Minnesota Ph.D. dissertation. 2022. Major: Computer Science. Advisor: Tian He. 1 computer file (PDF); 120 pages.With wider and deeper interaction between humans and systems in modern society, the study of human-centered cyber-physical systems (human-centered CPS) has become increasingly important. Thanks to the massive data collected by ubiquitous devices (e.g., smartphones) and advanced machine learning and data mining techniques, numerous human-centered CPS applications and studies are emerging. However, two essential problems still exist: (1) unlike purely internet-based systems, in human-centered CPS, different people engage the system at different places using different devices, which brings technical challenges like scalability and heterogeneity; (2) unlike CPS without wide and deep human participation, in human-centered CPS, human behavior (e.g., locations, mobility, activity) plays a key role, but human behavior is difficult to predict given its inherent uncertainty. To address the challenges, we have done a variety of works that can be organized under the three-layer framework of sensing, prediction, and decision-making. In the sensing layer, we design and build wireless sensing systems to capture human behavior like the arrival and departure at certain locations. We address the scalability challenge by studying human mobility and adopting their smartphones as virtual sensors, and we address the heterogeneity challenge by studying the impacts of environment and hardware on sensing and modeling the similarity with graph learning. In the prediction layer, we study the indoor localization problem by transforming it into a travel time prediction problem and solving it with graph learning based on the human behavior data collected from wireless sensing. In the decision-making layer, we utilize the data from the sensing and the knowledge from prediction to make decisions that lead to higher efficiency compared to the state-of-the-art. We also show how to utilize the feedback from humans to benefit the system design and achieve human-system synergy. In addition to the in-lab design and experiments, we implement our works in one of the latest and largest human-centered CPS applications, gig delivery. By studying couriers’ and merchants’ behavior and building corresponding sensing, prediction, and decision-making systems, we not only improve the system performance but also achieve the synergy between the couriers and systems, saving millions of dollars for the platform and benefiting millions and merchants, couriers, and customers.enBluetoothCyber-Physical SystemGig EconomyLocalizationMobile ComputingWireless SensingThesis or Dissertation